contrast and saturation in camera or software

Discussion in 'Digital SLR' started by mike regish, Apr 11, 2005.

  1. mike regish

    mike regish Guest

    Does it make any difference if I adjust the contrast and saturation in the
    camera settings or with software afterwards? If I don't like the results
    from camera settings, can they basically be reversed with software without
    getting any artifacts?

    TIA.

    mike
     
    mike regish, Apr 11, 2005
    #1
    1. Advertisements

  2. mike regish

    Steph Guest

    If you have the choice, do it yourself. Once done within the camera, it
    cannot be reversed and some tones will be lost forever. At least if you do
    it yourself, you can keep the original file unedited.

    That's why the output from semi-professional dslr looks rather dull and flat
    compared to a popular "amateur" model - no enhancement.
     
    Steph, Apr 11, 2005
    #2
    1. Advertisements

  3. mike regish

    Alan Browne Guest

    If you record RAW, then you can do whatever is needed after the fact.

    I don't believe contrast and saturation are reversible, at least at the
    extremes.

    I believe sharpness is reversible if you know how the sharp algo in the
    camera is set.

    In any case, the monitors on the camera are far too small to do any
    useful judgement of where these setting should be, IMO.

    Cheers,
    Alan

    --
    -- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
    -- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
    -- slr-systems FAQ project: http://tinyurl.com/6m9aw
    -- [SI] gallery & rulz: http://www.pbase.com/shootin
    -- e-meil: Remove FreeLunch.
     
    Alan Browne, Apr 11, 2005
    #3
  4. Nope. Sharpening is always irreversible.
     
    Ben Rosengart, Apr 11, 2005
    #4
  5. mike regish

    Mark Lauter Guest

    If you have the choice, do it yourself. Once done within the camera, it
    Exactly. Doing otherwise is like throwing the negative away after making
    the first print. :)
     
    Mark Lauter, Apr 11, 2005
    #5
  6. mike regish

    Jim Townsend Guest

    I think if you like what the camera is doing when you boost these
    parameters, then go with it. It will save you some editing time
    later on.

    To me, simple contrast adjustments aren't enough. Instead, I prefer
    to adjust levels and curves. These are far more powerful tools.

    I leave my camera settings in the neutral position and edit after
    the fact.
     
    Jim Townsend, Apr 11, 2005
    #6
  7. mike regish

    Alan Browne Guest

    Alan Browne, Apr 11, 2005
    #7
  8. To be honest, I can't readily explain why the process is
    irreversible, though it's obvious to me from the description
    of the algorithm on Luminous Landscape.

    http://www.luminous-landscape.com/tutorials/understanding-series/understanding-usm.shtml

    I can say that it discards information, but that's just restating
    the same thing in different words, and I won't insult your
    intelligence.

    Do you understand how sharpening works?
     
    Ben Rosengart, Apr 12, 2005
    #8
  9. mike regish

    Alan Browne Guest

    I have three (at least) sharpenning algorithms on two programs on my PC.
    For one of them, there is a 5*5 matrix of weights used when the
    matrix is passed over [I * F => I'] the image. Since on successive
    iterations of the filter (which I assume advances by 1 pixel at a time)
    would have a specific effect, then performing a reverse order pass with
    the filter set to F^-1 should result in the blurred back to orig. image.
    I think (eg: I'm not sure that information is discarded).

    Having said that, and not having simulated what I say above to verify
    it, and in no mood at all to try, I can't state whether it is so or not.

    Regarding USM filters (a la photoshop) which is what I use, I would not
    be surprised if it were not reversible due solely to the 'threshold'
    setting. From a USM'd image it would be impossible to determine which
    pixels were the result of a threshold that had been passed or not.

    I have no idea which algo. is in a camera, but I would guess it is
    similar to the first one I described as it is computationally light.

    Cheers,
    Alan

    --
    -- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
    -- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
    -- slr-systems FAQ project: http://tinyurl.com/6m9aw
    -- [SI] gallery & rulz: http://www.pbase.com/shootin
    -- e-meil: Remove FreeLunch.
     
    Alan Browne, Apr 12, 2005
    #9
  10. mike regish

    paul Guest


    No you cannot reverse these without loss of information but I believe it
    is best to let the camera do the adjustments if they are the appropriate
    adjustments because the camera should be using the raw data. If the
    scene is already too saturated, too contrasty or too noisy for
    sharpening, then it's better to turn that stuff off & do it yourself
    because highlights will be blown irretrievably, colors posterized, etc.
    I'm assuming RAW isn't an option for your camera or you don't have the
    energy to go through that for all shots. If you had the option to shoot
    RAW + high quality jpeg, you could probably ignore the RAW for most
    shots & use the jpeg. I would definitely boost the settings in that case.
     
    paul, Apr 12, 2005
    #10
  11. mike regish

    John Francis Guest

    Not true, in general.

    Plus, of course, if you're taking processed output from the camera,
    you've probably thrown away all the extra precision in the raw file.
     
    John Francis, Apr 12, 2005
    #11
  12. mike regish

    John Francis Guest

    Nope. A filter kernel is convolved with the image data, not multiplied.
    Convolving with the inverse of the weight matrix doesn't undo the effect.

    A very simple example: consider [[1 0 0] [0 1 0] [0 0 1]] as the filter.
    This (after scaling) replaces each pixel with the average of itself and
    it's two diagonal neigbours. That matrix is self-inverse, but applying
    the same filter again doesn't undo the effects.
     
    John Francis, Apr 12, 2005
    #12
  13. mike regish

    mike regish Guest

    I have raw as well as raw plus jpeg. Basically I was wondering if I could
    save tima at the computer by adjusting saturation and contrast in the
    camera, but from the answers here I'll leave the settings neutral.

    Thanks all.

    mike
     
    mike regish, Apr 12, 2005
    #13
  14. Hmm, ok, I was only familiar with PS-style USM sharpening. I don't
    know about this matrix algorithm you describe. (I don't "get it"
    from your description.)
    A USM shouldn't be too expensive as long as the radius is low ... I
    *think* ....
     
    Ben Rosengart, Apr 12, 2005
    #14
  15. mike regish

    paul Guest


    Then yes you can save time for most images if you have enough memory
    card capacity. If you have enough memory & not enough hard drive, delete
    the RAW files for exposures that look fine.
     
    paul, Apr 12, 2005
    #15
  16. mike regish

    John Francis Guest

    It's not really a matrix - it just happens to look a bit like one.

    It's a weight table for a discrete convolution filter; a two-dimensional
    array of weights. You generate the new value at a pixel by summing the
    products of the weights and the old value at the corresponding location,
    then you shift the weight array to a new position, sum the new set of
    products, and that's your new value for the next pixel. And so on.

    The value of the new pixel is unchanged if you add a new row or column
    of zeros to the weight table (and the table doesn't have to be square).
    Either of those possibilities makes it rather hard to find an inverse.
     
    John Francis, Apr 12, 2005
    #16
  17. mike regish

    Alan Browne Guest

    Doesn't mean I'm right either. Usually a matrix operation (of square
    matrices) can be reversed handilly with no error. I'm not sure of what
    kind of operation is used.
    Possibly. But as I said, in USM there is no way for you to know which
    pixels had been changed due to a threshold setting. So even if you know
    the weight and the radius, there would be no way to know whether a given
    pixel had been changed or not.

    Cheers,
    Alan.


    --
    -- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
    -- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
    -- slr-systems FAQ project: http://tinyurl.com/6m9aw
    -- [SI] gallery & rulz: http://www.pbase.com/shootin
    -- e-meil: Remove FreeLunch.
     
    Alan Browne, Apr 12, 2005
    #17
  18. mike regish

    Alan Browne Guest

    I forgot to add that I've found setting all settings to no change or
    neutral settings in the RAW import is the best way for me. Exception is
    the exposure setting. This way I do all the changes to the image at 16
    bit/color in PS, including (if needed) color, contrast, etc. and always
    USM for each o/p size.

    Cheers,
    Alan.

    --
    -- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
    -- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
    -- slr-systems FAQ project: http://tinyurl.com/6m9aw
    -- [SI] gallery & rulz: http://www.pbase.com/shootin
    -- e-meil: Remove FreeLunch.
     
    Alan Browne, Apr 12, 2005
    #18
  19. mike regish

    Alan Browne Guest

    If they're unchanged one way, then they'll be unchanged going back (and
    that is the case with one standard filter I have, the outer rows/cols
    are all 0, and the matrix is square. [5,5]). A non square matrix could
    be used. If that were the case, then inversing the operation would be
    an approximation.

    Cheers,
    Alan


    --
    -- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
    -- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
    -- slr-systems FAQ project: http://tinyurl.com/6m9aw
    -- [SI] gallery & rulz: http://www.pbase.com/shootin
    -- e-meil: Remove FreeLunch.
     
    Alan Browne, Apr 12, 2005
    #19
  20. mike regish

    John Francis Guest

    It's a convolution, not a matrix multiplication. Two very different things.
    (It's pretty easy to see that. In an n x n matrix multiplication, any one
    output value depends on only n multiplications of matrix values. A filter
    using a n x n table of weights uses each of those values to generate every
    single output value - n^2 cross-product terms, not just n).


    In an infinite-precision world, it is indeed be possible to calculate the
    reverse operation that would undo any particular filter[1]. But if you
    do that, you'll find a few things:

    o The reverse operation isn't confined to a small locale - the value
    at any pixel depends on the values at many post-filter pixels, at
    arbitrary large distances from the desired location.

    o The solution is often ill-conditioned; the coefficients can get
    arbitrarily large. This means that rounding and truncation errors
    in the original filter operation soon dominate the solution.

    o Even if you could somehow ignore those limitations, it's still
    a totally impractical approach for any reasonable-sized images.
    For a 6MP image, you'd end up solving a series of six million
    simultaneous linear equations.






    [1] You've got N (the # of pixels) simultaneous equations relating
    the N unknown original values to the N known post-filter values.
     
    John Francis, Apr 12, 2005
    #20
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.